[Doc] cuda and cudnn requirements for 1.18.1 gpu packages

[tianleiwu]

Description

Add CUDA and cuDNN requirements for 1.18.1 GPU package.

Motivation and Context

#21354
#21173

[jywu-msft]

isn't this a little misleading? this makes it look like our default packages are built with CUDA 12.x which is not the case.
I think until that becomes the default, we should list both CUDA 11.8/CuDNN 8.x and CUDA 12.x/CuDNN 9.x as the requirements and call this out explicitly.
@yf711 is looking into changing the defaults for ORT 1.19 (make the pypi/nuget default to CUDA 12.x and host the 11.8 packages on the alt feed)

[tianleiwu]

isn't this a little misleading? this makes it look like our default packages are built with CUDA 12.x which is not the case. I think until that becomes the default, we should list both CUDA 11.8/CuDNN 8.x and CUDA 12.x/CuDNN 9.x as the requirements and call this out explicitly. @yf711 is looking into changing the defaults for ORT 1.19 (make the pypi/nuget default to CUDA 12.x and host the 11.8 packages on the alt feed)

Installation is described in another doc, there are different ways for installing cuda 11.x and cuda 12.x:
https://onnxruntime.ai/docs/install/#install-onnx-runtime-gpu-cuda-11x
https://onnxruntime.ai/docs/install/#install-onnx-runtime-gpu-cuda-12x
That part will need change for next release 1.19.

Here we describe cuda and cudnn vesion. For example, if user has cuda 12, they will look up the first table to find cudnn version. If they have cuda 11, they need look at the second table to find cudnn version.

[jywu-msft]

isn't this a little misleading? this makes it look like our default packages are built with CUDA 12.x which is not the case. I think until that becomes the default, we should list both CUDA 11.8/CuDNN 8.x and CUDA 12.x/CuDNN 9.x as the requirements and call this out explicitly. @yf711 is looking into changing the defaults for ORT 1.19 (make the pypi/nuget default to CUDA 12.x and host the 11.8 packages on the alt feed)

Installation is described in another doc, there are different ways for installing cuda 11.x and cuda 12.x: https://onnxruntime.ai/docs/install/#install-onnx-runtime-gpu-cuda-11x https://onnxruntime.ai/docs/install/#install-onnx-runtime-gpu-cuda-12x That part will need change for next release 1.19.

Here we describe cuda and cudnn vesion. For example, if user has cuda 12, they will look up the first table to find cudnn version. If they have cuda 11, they need look at the second table to find cudnn version.

Ok yeah understood. your modifications improves on the previous docs. thanks! it's just that people might be expecting CUDA 12.x to be the default since it's listed first and they don't go and read the additional instructions. In any case, I think we need to switch CUDA 12.x to be the default for our package installations asap.